Israel-based proteanTecs provides deep level visibility into chips for continuous monitoring and predictive maintenance.
AI, edge computing, 5G and the Internet of Things (IoT) are among the key transformative technologies pushing the limit of today’s service-oriented markets. These trends are offering users absolute availability, mega-functionality and instantaneous access to data, seemingly effortlessly, bringing new opportunities to brand owners in the datacenter, communications and automotive industries. Service providers rely on increased flexibility, resilience, and greater visibility into their supply chains and business activities. New capabilities, such as remote monitoring and predictive maintenance, provide an exciting new opportunity to maintain serviceability and cost at scale. This need has been especially true during the recent global pandemic, which has crippled traditional methods.
That is a “macro” level approach, if you will. But one startup has taken the same concept into the sub-micron level to provide deep visibility into what’s happening inside advanced electronics by making the chips inside smart. With this first of its kind technology, manufacturers and service providers can gain insight into the systems’ health and performance during the production process and while the system is deployed to its mission in the field.
“Over the life of the electronics, you’ll have degradation, you’ll have changes in software and other types of things that affect the system. The ability to see inside the chip throughout its lifetime, is becoming a very important thing,” explains Keith Morton, CRO at proteanTecs, a four-year old Israel-based start-up that has developed deep data analytics, combining a cloud-based platform that reads out data from on-chip “agents”, and machine learning, to predict faults before they become failures.
So, what exactly is proteanTecs trying to address?
“If you think about electronics today, in almost every area in your life, they are getting more complex and their scale is just exploding,” says Morton. “Think about a data center. With the amount of hardware —tens of thousands of systems and enormous amounts of processors—preventing IT system failures or being able to control the performance of the system is a very daunting task. Or in a 5G base station, you have tons of electronics packed in a small space, the scale of which makes it very difficult to maintain reliably and push the power-performance envelope.”
proteanTecs uses a system’s central processing unit to understand the impact of silicon wearout, application stress or environmental effects. “We employ Universal Chip Telemetry (UCT) by embedding agents into the chips during design. The agents are essentially small monitoring IPs that look at parametric phenomena at a very high coverage, while the system is operating. This allows manufacturers and brand owners to continuously monitor what’s happening inside the chip, at a very fine level of detail, throughout its entire lifetime,” says Morton. “It is a unique and very comprehensive solution—and right now, it is an exciting new category. These agents that go into the chip—like “eyes” inside—are actually built for analytics. The real magic is the platform – an application runs on the cloud, reading data from inside of the chip and processing that data, with machine learning. This is done at production testing and in the field, after you’ve deployed those systems. The platform can also integrate additional sources of data, from customer databases or 3rd party platforms, to enhance the insights.
Since the agents are specialty monitors, tailored for analytics, proteanTecs provides a suite of EDA tools to determine where to place the agents and how to integrate them automatically. The agents work together to provide a full picture of the chip’s health and performance on multiple dimensions.
Deep Data visibility, from within.
Some companies may cobble together existing technologies to help specific pieces of the value chain better understand their design, manufacturing, or in-field processes. But, without a holistic solution, data silos get reinforced and companies struggle to correlate data in a common “language”, even amongst internal teams.
“We generate novel data throughout the lifecycle, based on analysis of the same UCT measurements at every stage. The thing that’s unique is that our IP and our analysis software were all developed together, with the target solution in mind. In our case, we built the entire system holistically—so when we design the agents, we are already thinking about the analytics, what kind of data we need to extract. The technology was developed by teams of experts in the fields of chip design, data science and cloud software working together. As a startup, we had the luxury of starting with a blank sheet of paper, and thinking about the IP, the analysis, the integration, how to get the data out of the chip, and what to do with the data,” Morton explains.
And rightfully so, because proteanTecs has a strong hardware, software, and systems background. Its DNA comes from a very successful electronics company called Mellanox (acquired by Nvidia), which three of the proteanTecs founders also founded. According to Morton, what’s exciting about proteanTecs’ founders being from Mellanox is that not only did they develop chips with in-house production capabilities, but they also made their own systems and software.
“They had a lot of expertise across the entire value chain—they weren’t just making chips, they were also selling systems. So, they really understood all the complexities, and that’s what gave them the vision for this company—to provide the tools that they always wished they had when they were heading Mellanox,” says Morton.
“UCT breaks down the barriers between designers, product engineers, system engineers, and software and maintenance teams, because now we have common data on a single analytics platform, from design to the field”, says Morton. “Different teams seek out different insights: test engineers might see how the a certain application is stressing the chip; designers might see if there is an issue with the performance margins; Q&R engineers might see outliers so they can reduce DPPM [defective parts per million], and field operations teams might see performance degradation over time,” says Morton.
Embedding UCT agents means proteanTecs must work closely with customers. “From day one in the company, even before we had a product, our CEO and our CTO set clear demands that we hold true to, such as: we cannot affect the area of the chip, we cannot affect the performance of the chip, and we cannot affect power in a significant way, because otherwise, the cost of adopting it is too high,” explains Morton.
So far, the company’s solution has been proven on many chips, in advanced process technologies, from 28nm down to 7nm and 5nm nodes. “At that technology, we’ve never had an increase in the size of the block, we’ve never had to compromise on performance, and any power that we’ve added is very minor and something that could be managed,” says Morton.
Some of the company’s customers are in the Fortune 100, and even Fortune 10 companies. “We are working with the leading company in the cloud computing area, and we have multiple hyperscalers deploying our technology in advanced chips,” says Morton. “We also have communications—hardware and networking customers that are the biggest companies in the world doing large switches. We also have multiple storage customers that are working on commercial grade storage in advanced technologies.”
proteanTecs is very committed to the automotive space. “I think it’s a really exciting time; things are moving really fast and we are already seeing a lot of autonomy today. When you look at self-parking and automatic braking, and other types of things. It’s coming in gradually, but I think even faster than anyone expected. With some companies, like Tesla, taking a very aggressive approach and putting a lot of advanced electronics in a car, and then everybody else investing in this area, it’s progressing quickly —but it’s something that we have to do right: you can’t afford to put so much hardware redundancy in a car where the cost becomes too high,” says Morton.
Through its technology, proteanTecs is providing much more data so that manufacturers can reduce the amount of redundancy, while still ensuring safety. “Our solution is very tightly coupled with the whole functional safety area, as well as maintaining the reliability of the car,” explains Morton. “For instance, we can generate alerts from the car – we can send an alert to turn off the car, or calculate the time to failure, so that potentially unsafe vehicles aren’t on the road. This allows OEMs to transition from preventive to predictive maintenance strategies. In other cases, we can signal that a vehicle needs an over-the-air update to address a specific issue. Whenever you have a car that’s being driven by a computer, you must continuously update that software and make changes to it. When you update that software, it’s very difficult to know how the electronics are going to be affected, especially since they are in the car. In our case, we can see what’s happening in the electronics and see how they are reacting to the new software. This over-the-air update and understanding how that impacts the car is huge.”
The company recently opened an office in Taiwan to meet the rapidly increasing market demand in the Asia Pacific region. It chose Taiwan because of its being the epicenter of the electronics industry in the region.
“For us, it is a very easy decision to start in Taiwan. You have best-in-class manufacturing; you have some of the top fabless companies in the world; you have some of the best packaging and test houses in the world; and then you have a lot of research agencies here. This is all paired together with top universities who prepare the next generation of incredible talent,” says Morton. “To us, that’s a recipe for success. All the right ingredients are in one place, in a connected community, where you can make things happen.”
The platform is at the heart of what proteanTecs is doing. “We’re a data company, and our solution is designed to provide visibility to the electronics industry, so it can scale reliably and efficiently,” says Morton.
“We’re really focused on the platform and the lifecycle; in particular, predictive maintenance. Being able to predict maintenance needs of complex electronics, being able to detect faults before they become catastrophic failures, to generate alerts, and to allow performance improvements in the field. Those are all the key things that we do. Optimization from design to the field allows us to be the best,” he concludes.